BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Robust multilingual syntactic parsing - Ryan McDonald\, Google
DTSTART:20140530T110000Z
DTEND:20140530T120000Z
UID:TALK50785@talks.cam.ac.uk
CONTACT:Tamara Polajnar
DESCRIPTION:In this talk I will present two advances in graph-based depend
 ency\nparsing. In the supervised learning setting\, I will present a simpl
 e\ncube-pruning framework that allows for the scoring of arbitrary\nhigher
 -order features with little loss in accuracy over exact methods.\nCritical
  to its success is a generalization of the max-violation\ntraining strateg
 y of Huang et al. (2012) to approximate search over\nhypergraphs -- of whi
 ch cube-pruned dependency parsing is a special\ncase. Experiments on a num
 ber of languages shows consistent\nimprovements over other state-of-the ar
 t parsers with speeds\ncomparable to linear transition-based parsers. Next
 \, I will\ninvestigate how to adapt graph-based parsers to a specific targ
 et\nlanguage in a cross-lingual learning setting. Specifically\, I will\ns
 how how features derived from the World Atlas of Language coupled\nwith a 
 novel 'ambiguity-aware' self-training algorithm can lead to\nlarge improve
 ments in accuracy\, especially for target languages that\nare typologicall
 y diverse from the set of source languages. The talk\nwill conclude with s
 ome general opinions on the future of dependency\nparsing.\n\nThis is join
 t work with Hao Zhang (Google)\, Oscar Täckström (Google)\,\nJoakim Nivr
 e (Uppsala)\, Liang Huang (CUNY) and Kai Zhao (CUNY).
LOCATION:FW26\, Computer Laboratory
END:VEVENT
END:VCALENDAR